DeepViewRT™ Inference Engine
Best In Class Performance and Unprecedented Portability
The DeepViewRT runtime inference engine provides developers with the freedom to quickly deploy ML models to a broad selection of embedded devices and compute architectures without sacrificing flexibility or performance.
Select a proven public model from DeepView model zoo, create or convert your own model with NXP's eIQ portal and compare performance tradeoffs between quantized and floating point under real-world runtime conditions.
Lastest DeepViewRT Benchmarks
Real world inferences (and counting)
The DeepViewRT engine has been highly optimized for runtime size and performance across a long list of the most popular embedded processors, architectures and standard x86 class devices - this means you can run public, custom and proprietary ML models anywhere the DeepViewRT engine is supported.
Best of all, it's FREE for development and production.
Benefits of the DeepViewRT production-ready engine
Tested & documented for quick out-of-the-box deployment
Examples and tutorials to save you time getting started
Field proven to avoid surprises when you ship your products
Lifecycle management for stability, longevity & compatibility
Professional support if you need it
Embedded: Linux, Android, Azure, FreeRTOS and bare metal
Desktop: Linux and Windows
Processor Types & Compute Architectures:
Microcontrollers (MCPUs): Arm Cortex M7
Application Processors(CPUs): Arm Cortex A35, 53, 72
Graphics Processing Units (GPUs): OpenVx
Neural Processing Units (NPU's): VeriSilicon and Ethos*
Desktop: x86 (development & validation)
Model deployment formats:
Floating point for full precision accuracy
Fixed point for optimal size and efficiency